编码代理中恶意技能文件的风险评估研究
原标题：Towards a Risk Assessment of Malicious Skill Files in Coding Agents
AI 摘要
该研究针对编码代理中的恶意技能文件提出风险评估方法，通过六种LLM将471个真实shell命令转化为2826个看似良性的技能基准，并评估了Gemini CLI和Qwen Code两个企业级代理，发现Gemini CLI在95.5-96.1%的运行中被利用，Qwen Code在71.6-74.0%中被利用，表明技能接口存在严重安全风险，企业需在采用前进行缓解。
正文节选
Computer Science > Software Engineering Title:Towards a Risk Assessment of Malicious Skill Files in Coding Agents View PDF HTML (experimental) Abstract:Autonomous coding agents are increasingly embedded in enterprise software workflows with delegated authority over connected systems. Central to this architecture is the agent skills interface: folders of instructions and scripts that agents load dynamically to specialize their behavior. This interface also widens the attack surface, l